飞行器集群敌我识别是集群协同作战与安全控制的关键，是实现可靠自主决策与安全任务执行的重要保障。目前的敌我识别方法大多依赖雷达特征、通信标识或专用识别设备，普遍存在部署成本高、抗干扰能力弱、可解释性欠缺等不足。为此，提出了一种基于运动拓扑推理的敌我识别新方法SRIFF(Swarm Relational Inference for Friend or Foe)。首先，构建了融合群体动力学规则与图结构建模的集群可解释特征模型，实现了敌我识别过程和依据的物理可解释性；其次，设计了面向可变规模集群的自适应时序编码机制，实现了对集群动态行为特征的统一表征与有效提取；最后，采用变分推理框架下的双解码器协同优化方法，实现了拓扑推理、轨迹预测与敌我识别多任务约束下的端到端联合优化效果。仿真结果表明，SRIFF方法具备复杂干扰与规模变化条件下的稳定识别能力，达到了优于现有方法的识别精度与性能表现，具有为安全敏感场景提供高效、低成本、可解释敌我识别方案的作用。

Identification of Friend or Foe (IFF) for aircraft swarms is a critical prerequisite for cooperative operations and safe autonomous control, providing essential support for reliable decision-making and secure mission execution. Existing IFF approaches largely rely on radar signatures, communication identifiers, or dedicated identification devices, which often suffer from high deployment costs, vulnerability to interference, and limited interpretability. To address these limitations, this paper proposes a novel motion topology inference–based IFF method termed SRIFF (Swarm Relational Inference for Friend or Foe). First, an interpretable swarm feature modeling framework is constructed by integrating collective dynamics principles with graph-based representations, enabling physically interpretable reasoning of the IFF process and its underlying decision basis. Second, an adaptive temporal encoding mechanism is designed to accommodate variable-scale swarms, allowing unified representation and effective extraction of dynamic behavioral features. Finally, a variational inference framework with dual-decoder cooperative optimization is introduced to achieve end-to-end joint learning of topology inference, trajectory prediction, and IFF under multi-task constraints. Simulation results demonstrate that SRIFF exhibits stable and robust identification performance under complex interference and swarm scale variations, achieving superior accuracy and overall performance compared to existing methods. These results indicate that SRIFF provides an efficient, low-cost, and interpretable IFF solution for safety-critical swarm applications.